Automated detection of marine animals using multispectral imaging

Personnel-based aerial surveys have demonstrated their capability for monitoring marine animal populations. That being said, not many studies have been conducted on smaller shark species in coastal waters, where they can become a threat to beach-goers and possibly be killed by beach protection netting. In this paper we report our findings on real-time, automated detection of great white shark dummy targets using a multispectral imager. 2.5 × 1.2 m plywood shark cutouts were submerged to 1 m, 2 m, and 3 m below the surface in the Pacific Ocean off the coast of San Diego, CA. 66 passes were conducted over the course of three days, resulting in an average probability of detection of 84.8%. Automatic detection was mostly limited by water turbidity in this test, but other environmental conditions such as wind, wave speed and period, and sun glint also had an effect on detection. The high probability of detection and low false alarm rate from this study are an indication that using an automated remote sensing system could be an effective alternative to personnel-based aerial surveys.